Prof Joseph Davis presented "Human Computation and Crowdsourcing: Conceptual Foundations, Emerging Directions" at The Hong Kong Polytechnic University on 8 December 2014.
Abstract: Harnessing large-scale human computation through crowdsourcing and its integration with machine computation has emerged as an important research focus in recent years. I will present the conceptual foundations of crowdsourcing and develop a preliminary taxonomy of crowdsourcing applications. My primary focus will be on the integration of machine and human computation for solving complex problems such as protein folding and image processing. I will
also describe the design of a hybrid machine-human system for image similarity search which uses the machine to perform the heavily computational task of selecting a set of ranked, similar images for a given query image, based on well-known algorithms such as SIFT, SURF, SURF128, and ORB. The humans-in-the-loop accessed through the Mechanical Turk platform performed the similarity-based rank ordering of the selected images generated by each of the four algorithms. An experimental comparison of the performance of the machine-only ranking with the ranking provided by the hybrid system (against the gold standard using Corel-Princeton image databases) showed that the latter achieved statistically significant performance